> time forecasting is bullshit

for a model to be useful, it doesnt need to capture the behavior of a system. It only needs to capture signals which can be useful. For example, for a biased coin toss, a model is already useful if it can predict a little better than random.